ml-agents/ml-agents/mlagents/trainers/bc/online_trainer.py

# # Unity ML-Agents Toolkit
# ## ML-Agent Learning (Behavioral Cloning)
# Contains an implementation of Behavioral Cloning Algorithm

import logging
import numpy as np

from mlagents.envs.brain import AllBrainInfo
from mlagents.envs.action_info import ActionInfoOutputs
from mlagents.trainers.bc.trainer import BCTrainer

logger = logging.getLogger("mlagents.trainers")


class OnlineBCTrainer(BCTrainer):
    """The OnlineBCTrainer is an implementation of Online Behavioral Cloning."""

    def __init__(self, brain, trainer_parameters, training, load, seed, run_id):
        """
        Responsible for collecting experiences and training PPO model.
        :param  trainer_parameters: The parameters for the trainer (dictionary).
        :param training: Whether the trainer is set for training.
        :param load: Whether the model should be loaded.
        :param seed: The seed the model will be initialized with
        :param run_id: The identifier of the current run
        """
        super(OnlineBCTrainer, self).__init__(
            brain, trainer_parameters, training, load, seed, run_id
        )

        self.param_keys = [
            "brain_to_imitate",
            "batch_size",
            "time_horizon",
            "summary_freq",
            "max_steps",
            "batches_per_epoch",
            "use_recurrent",
            "hidden_units",
            "learning_rate",
            "num_layers",
            "sequence_length",
            "memory_size",
            "model_path",
        ]

        self.check_param_keys()
        self.brain_to_imitate = trainer_parameters["brain_to_imitate"]
        self.batches_per_epoch = trainer_parameters["batches_per_epoch"]
        self.n_sequences = max(
            int(trainer_parameters["batch_size"] / self.policy.sequence_length), 1
        )

    def add_experiences(
        self,
        curr_info: AllBrainInfo,
        next_info: AllBrainInfo,
        take_action_outputs: ActionInfoOutputs,
    ) -> None:
        """
        Adds experiences to each agent's experience history.
        :param curr_info: Current AllBrainInfo (Dictionary of all current brains and corresponding BrainInfo).
        :param next_info: Next AllBrainInfo (Dictionary of all current brains and corresponding BrainInfo).
        :param take_action_outputs: The outputs of the take action method.
        """

        # Used to collect teacher experience into training buffer
        info_teacher = curr_info[self.brain_to_imitate]
        next_info_teacher = next_info[self.brain_to_imitate]
        for agent_id in info_teacher.agents:
            self.demonstration_buffer[agent_id].last_brain_info = info_teacher

        for agent_id in next_info_teacher.agents:
            stored_info_teacher = self.demonstration_buffer[agent_id].last_brain_info
            if stored_info_teacher is None:
                continue
            else:
                idx = stored_info_teacher.agents.index(agent_id)
                next_idx = next_info_teacher.agents.index(agent_id)
                if stored_info_teacher.text_observations[idx] != "":
                    info_teacher_record, info_teacher_reset = (
                        stored_info_teacher.text_observations[idx].lower().split(",")
                    )
                    next_info_teacher_record, next_info_teacher_reset = (
                        next_info_teacher.text_observations[idx].lower().split(",")
                    )
                    if next_info_teacher_reset == "true":
                        self.demonstration_buffer.reset_update_buffer()
                else:
                    info_teacher_record, next_info_teacher_record = "true", "true"
                if info_teacher_record == "true" and next_info_teacher_record == "true":
                    if not stored_info_teacher.local_done[idx]:
                        for i in range(self.policy.vis_obs_size):
                            self.demonstration_buffer[agent_id][
                                "visual_obs%d" % i
                            ].append(stored_info_teacher.visual_observations[i][idx])
                        if self.policy.use_vec_obs:
                            self.demonstration_buffer[agent_id]["vector_obs"].append(
                                stored_info_teacher.vector_observations[idx]
                            )
                        if self.policy.use_recurrent:
                            if stored_info_teacher.memories.shape[1] == 0:
                                stored_info_teacher.memories = np.zeros(
                                    (
                                        len(stored_info_teacher.agents),
                                        self.policy.m_size,
                                    )
                                )
                            self.demonstration_buffer[agent_id]["memory"].append(
                                stored_info_teacher.memories[idx]
                            )
                        self.demonstration_buffer[agent_id]["actions"].append(
                            next_info_teacher.previous_vector_actions[next_idx]
                        )

        super(OnlineBCTrainer, self).add_experiences(
            curr_info, next_info, take_action_outputs
        )

    def process_experiences(
        self, current_info: AllBrainInfo, next_info: AllBrainInfo
    ) -> None:
        """
        Checks agent histories for processing condition, and processes them as necessary.
        Processing involves calculating value and advantage targets for model updating step.
        :param current_info: Current AllBrainInfo
        :param next_info: Next AllBrainInfo
        """
        info_teacher = next_info[self.brain_to_imitate]
        for l in range(len(info_teacher.agents)):
            teacher_action_list = len(
                self.demonstration_buffer[info_teacher.agents[l]]["actions"]
            )
            horizon_reached = (
                teacher_action_list > self.trainer_parameters["time_horizon"]
            )
            teacher_filled = (
                len(self.demonstration_buffer[info_teacher.agents[l]]["actions"]) > 0
            )
            if (info_teacher.local_done[l] or horizon_reached) and teacher_filled:
                agent_id = info_teacher.agents[l]
                self.demonstration_buffer.append_update_buffer(
                    agent_id,
                    batch_size=None,
                    training_length=self.policy.sequence_length,
                )
                self.demonstration_buffer[agent_id].reset_agent()

        super(OnlineBCTrainer, self).process_experiences(current_info, next_info)